How to explain one science idea in four shots

Start with the mistake a viewer might make

Put a glass of ice water on a dry counter and ask where the drops on its outside come from. A short video can answer with a close-up, a few words, and a comparison. It can also teach the opposite lesson by showing liquid seeping through the glass. That single wrong image can undo a correct sentence in the narration.

On November 5, Google announced Gemini Omni 1.2 Flash with native video generation and editing up to 15 seconds. Google says users can request durations from three to fifteen seconds in one-second steps, supply references, and request changes in follow-up turns. The announcement describes the tool’s capabilities. It does not show that a science explanation will keep the right cause, timing, or label. The four-shot plan here is a production example, not a report of testing the model.

Start your own plan by writing the incorrect answer you want a viewer to discard. For this one: “The cold water escaped through the glass.” Then write the explanation you want them to leave with: water vapor in the nearby air becomes liquid on a sufficiently cold outer surface. The U.S. Geological Survey uses a cold glass to illustrate condensation, and a NOAA teaching activity connects the effect to nearby air cooling below its dew point. You do not need a lecture on humidity for the clip. You do need to keep the outside surface visible and account for the water without implying that it crosses the glass.

Choose one audience question before drafting any prompt. “Where did the drops come from?” can fit this plan. “How does condensation shape Earth’s water cycle?” needs different examples and more time. A useful fifteen seconds settles one confusion well enough that viewers know which longer question to ask next.

Source: Google: Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips ↗ · USGS: Condensation and the Water Cycle ↗ · NOAA: Learning Lesson: Sweatin’ to the Coldies ↗

Give each shot a job and a stopping point

Seconds zero through three establish the question. Show two matching transparent glasses side by side on the same dry surface. Put ice water in the left glass and room-temperature water in the right. Frame both bases and the outside walls, and keep the background plain. Begin with dry exteriors in the illustration. The viewer should have time to register two glasses, one with ice, before the camera moves. Put the words “Ice water” and “Room-temperature water” near their respective glasses in the edit.

Seconds three through seven show the difference. Move in toward the outside wall of the cold glass while keeping its recognizable rim or base in view. Use a cut that indicates time has passed: the wall is dry at the start of the story and has droplets later. Let the editor add the words “Later” rather than inventing a countdown. The clip’s length is not a measurement of how quickly droplets appear. Avoid a stream of water entering from above or a wet counter, either of which can suggest a spill instead.

Seconds seven through eleven explain the cause in a deliberately simpler visual style. A clean diagram can place the words “Water vapor in air” outside a line representing the glass wall and point toward a drop on the outer surface. Keep the arrow on the air side of the line. If its head ends inside the glass, it sends the wrong message regardless of what the voice says. A visible diagram caption makes the change from an illustration of the setup to an explanation of an invisible process easier to follow.

Seconds eleven through fifteen return to the paired view. Leave the droplets on the left glass and the right glass comparatively dry in the depicted conditions. End with “Drops came from the air” rather than adding another new concept. The return to the original framing gives the audience a chance to compare the two surfaces and ties the explanation back to the question. A final hold also gives viewers who are reading rather than listening time to finish.

Keep a list of what cannot change

Write a small continuity record for the setup. The left glass is cold, the right glass is the comparison, their shapes match, both sit on the same counter, and the camera stays on the same side of the arrangement. The liquid levels should stay plausibly consistent across the brief sequence. Most importantly, every visible condensation droplet is on the exterior of the cold glass. Those details give reviewers something to check. “Consistent and realistic” does not tell them which inconsistencies would change the explanation.

For each cut, note what carries the viewer across it. When a wide shot becomes a close-up, an identifiable rim and the position of the ice help connect the close-up to the left glass. When the close-up becomes a diagram, keep the outer face on the same side or label the orientation plainly. When the diagram returns to the wide shot, the two glasses should be in their starting positions. If a generated edit flips the setup, it can make the first labels apply to the wrong glass even if the new shot is attractive on its own.

A reference image can describe the setup, and Google says the model accepts reference inputs. Still, keep editorial constraints independent of the tool. Ask for a stable view and a specific change, then inspect the resulting frames rather than relying on the request as proof. If a generation puts condensation on both glasses or changes which glass contains ice, the draft has failed its instructional comparison. A correctly filmed or illustrated still of the two glasses can be more useful than a more dynamic shot with the wrong relationship.

The comparison is conditional. In real life, the temperature of each surface and the air’s moisture affect whether droplets form. The right glass is a comparison for the stated setup, not a claim that a glass without ice never collects condensation. Save that qualification for a description or companion text if it would overload the frame.

Write the labels outside the generator

Make a tiny label list before production. For this story it could contain “Ice water,” “Room-temperature water,” “Later,” “Water vapor in air,” “Condensation,” and “Drops came from the air.” Each label has one home. The two glass labels establish the setup, the time label qualifies the cut, and the diagram words explain the source and process. They do not all need to appear at once. If the ending repeats information that already reads clearly, use the spare time to hold the comparison.

Add the text in a video editor so it remains editable and can be checked against the intended wording. Place each label where its connection is unambiguous. For example, attach “Condensation” to an outside drop or its immediate neighborhood, rather than setting it over the ice cubes. Check that captions do not obscure the lower parts of the glasses and that small screens preserve the relationship between words and objects. If you crop for a vertical version, recheck the edit; both glasses and both labels must still be present when the audience needs the comparison.

One possible narration is: “Why is the cold glass wet outside? Air near it cools. Some water vapor turns into liquid on the glass. The drops came from the air.” Read it aloud with the four beats. If it feels rushed, remove a repeated phrase or let a clear image carry some of the meaning. Do not speak a label about vapor while showing only ice; place the line where the diagram establishes the air side. Caption the spoken words separately from any labels that identify the visuals.

The scientific word has a specific job. “Water vapor” names the gaseous form of water, which the diagram can represent with text. It should not be drawn as a puff of visible smoke pouring out of the glass. A neutral schematic makes it easier to describe what the audience cannot see directly without creating a new false observation.

Mark illustration, demonstration, and measurement

A photographic look can make a generated glass seem like footage of a classroom demonstration. Give viewers a visible way to tell what it is. Use a short on-screen note such as “AI-generated illustration” when the visual is generated, and identify the schematic as a diagram. If you use real footage as well, label the change between illustration and footage and say what each segment is doing. A credit at the end may not travel with a clip trimmed for another platform.

If you want the audience to see an observation, film an actual setup and describe the relevant conditions you know. A shot of drops on a glass is an observation of drops under those conditions. It does not by itself show where the water molecules came from. The explanation comes from the physics and can be supported by a comparison, a source, or a separate investigation. The same difference matters in more consequential stories: an illustration of a cloud forming is not weather evidence, and a rendered instrument readout is not a measurement.

Avoid adding simulated precision that looks like data. This example has no measured temperature, humidity, or onset time, so the storyboard should not display numbers for them. If a production includes time compression in actual footage, label it and keep the timing record if the rate matters. If the only timing information is that the drops appear in a later shot, “Later” is enough. A viewer can learn the direction of the change without being given a fabricated value.

When the clip accompanies an article, include a short description that names the phenomenon and links to the teaching references. A teacher or viewer should be able to ask how the explanation is supported even if they do not know which generator made the images.

Review the cut against the question

Make the first review about causality. Freeze the frame before and after each cut and ask what changed. You should be able to name the change in a sentence: two initially dry exteriors; droplets later on the cold glass; an explanation on the air side; the same glasses after the change. If you have to explain away a disappearing glass, moving label, or jump in the fill level, decide whether it distracts or actually supplies the wrong cause. Fix the second kind before worrying about the beauty of the cut.

Then ask a fresh viewer to watch once and say where the drops came from and why the right glass was shown. Avoid repeating the right answer in the question. If they say the drops melted off the ice or passed through the wall, find the shot that made that interpretation plausible. A cut from ice to exterior drops might invite that reading even if the narration is accurate. Giving the diagram more time may do more than adding a faster sentence.

Review the actual exported clip muted as well as with sound. The labels and captions should preserve the causal explanation in silence, and the words should refer to the same parts of the frame with sound on. Scrub through the diagram to ensure the arrow stays outside the wall throughout its appearance. Watch the last second for text that disappears too soon or a disclosure lost to an end card. These checks focus the review on what a viewer can recover from the final file, rather than what the storyboard promised.

If the generated motion does not hold up, narrow its role. Use it for an opening illustration, keep a human-authored diagram for the cause, or replace the disputed shot with an actual recording. The production choice follows the lesson: every cut should help viewers tell what they saw, what explains it, and which parts of the clip were made to illustrate that explanation.